How to identify residency via SVF?

79 views
Skip to first unread message

yu lei

unread,
Mar 25, 2022, 8:49:07 AMMar 25
to ctmm R user group
Hi Chris

 Thanks so much for your reply last time. 

 I know the AKDE is suited for  range residents, and we can determine whether an individual exhibit range residency via visual SVF with reaching asymptotes  or the DOF area value.
       However, I'm not sure when the variogram asympetote can indicate residency.  Recently,  i saw a figure (posted below ) in the preprint paper (https://osf.io/ka5t6) , it seems stable variograms have kinds of shapes, including bell, hump shapes, even some consistent upward lines which should not be stable events i think.  
       I only suppose that when the vairogram have a flatten asymptote (like a straight line) can represents range residency before,  After saw the figure, i was confused about this question. 
 Thanks for any suggestions.
 Best,
Yu Lei
variogram.jpg

Christen Fleming

unread,
Mar 28, 2022, 12:30:03 AMMar 28
to ctmm R user group
Hi Yu Lei,

The ideal variogram of a range resident individual is one that has a flat asymptote, like F04 & M03 in the figure. However, I think that these variograms are all fully zoomed out, and the last half of a variogram is pretty erroneous, so variograms like F01 & F15 are okay too.
The highlighted variograms F08, F10, F12 are the most problematic and definitely indicate something non-resident happening.
The middling variograms, like F03 and F06 are somewhat questionable, and I would look at whether their area estimates and DOFs fall within the spread of the population.

We have a number of features to help with this identification, including the CI="Gauss" argument to variogram() and the cluster() function, and we are testing some cross validation methods as well.

Best,
Chris

Jesse Alston

unread,
Mar 29, 2022, 11:35:18 AMMar 29
to ctmm R user group
Hi Chris,

To build on Yu Lei's question, I've been trying to think about what the shape of the variogram will do to predictions of home range use. Consider half-sample cross-validation like was performed in Noonan et al. 2019 ( https://doi.org/10.1002/ecm.1344 ). If a variogram continues to go up throughout the timeline on the x-axis, a home range estimated using the first half of a data set will consistently and greatly underestimate the number of points within that home range estimate in the second half of the data set. I think that's pretty clear. If a variogram goes down toward the end of the timeline, though, I don't think you'll get the opposite: consistent overestimation of the number of points in the test set within the home range. Is that right? I might look into this with empirical data.

Jesse
Message has been deleted

yu lei

unread,
Mar 30, 2022, 9:21:48 PMMar 30
to ctmm R user group
Hi Chris,
Thanks for your prompt reply.  I have another two questions here.
First, i find the GUESS i manually fitted (red line in below ) is different with the fitted GUESS of best model via ctmm.select (green line). Should i according to the fitted model result to estimate home range or adjusting the model GUESS until close i wished. If necessary to adjust, how to realize?
BFUD002_2B.jpgBFUD002_2B_final.jpg

The second question is about movement parameters from ctmm.select function (see below). 
model results.jpg
Two parameter (home range crossing time (τp )and directional persistence timescale (τv )) were referred in this paper  Medici 2022 (https://doi.org/10.1186/s40462-022-00313-w),  does it mean τ position and τ velocity separately ? 
And i confuesd on diffusion parameter which have rare interpretation before, it means movement or dispersal range per day ? if so , can i use it to quantify the movement or space use of animal individual during study period?
Thanks for any suggestions.
Best,
Yu Lei

Christen Fleming

unread,
Mar 31, 2022, 9:19:29 AMMar 31
to ctmm R user group
Hi Jesse,

One thing to consider is that the x-axis of the variogram isn't the timeline/timestamps of the data, but different sized time lags between data points, with the maximum time lag being the total period of the data (first-to-last point), but with lesser lags occuring at different times within the period of the data.

A variogram that continues to increase without limit (like the highlighted examples) will likely underestimate future space use with any estimate based on the current data, like you say. And if this is happening on a large timescale, then it's probably some kind of dispersal phenomena taking place.

A variogram that consists of a hump could be something more like an A -> B -> A movement. This could be two crossings of a large home range, or a dispersal and return behavior, or something else, depending on the context.

Best,
Chris

Christen Fleming

unread,
Mar 31, 2022, 9:26:17 AMMar 31
to ctmm R user group
Hi Yu Lei,

Seeing this large of a discrepancy between the empirical and theoretical (fitted) variograms suggests that there is some non-stationary behavior in the data that needs to be taken into account. I would look to see if the data needs to be segmented at some time(s) because of a range shift or dispersal event of some kind.

The tau parameters are as you have noted. The diffusion rate parameter is new and will be described in an upcoming paper. I introduced this for people to be able to quantify a rate of movement in data that are too coarse for speed estimation, but not so coarse that they are IID.

Best,
Chris
Reply all
Reply to author
Forward
0 new messages